Gemini 4 Carbon: Google’s strongest model, for cyber defenders only so far
The model for long engineering tasks topped the Artificial Analysis index. Developers and businesses must wait while Google tests its safeguards.
On October 28, two weeks after Argon, Google introduced Gemini 4 Carbon. The company calls it its strongest model for cybersecurity and long-horizon work.
What’s new
- Ranking: first on the Artificial Analysis Intelligence Index v5.0 with a score of 59.
- Long tasks: Google says internal teams saw it run for days, working through dead ends on multi-step bug-fixing tasks.
- Cyber defense: 88.2% on SEC-bench Pro (finding vulnerabilities) and 71% on CWE-bench (patching root causes).
- Safety: high-severity misalignment alerts in internal live coding fell by almost 80%, and unsupported claims by about half compared with Argon.
Who gets it
First, a subset of cyber defenders, then more trusted partners through the Fairwind Program. Google promises it to developers, businesses and consumers “soon”, after gathering feedback and refining safeguards. The company is also taking part in the US government’s voluntary pre-release testing.
Introductory pricing is $2 per million input and $10 per million output tokens, rising later to $4 and $20.
Limitations
Nearly all the numbers come from Google, and some rest on internal evaluations. There’s no firm date for wider access.
Prepared with AI from public sources.